9.4 Example Applications

9.4 Example Applications
9.4示例应用

Fig. 9.4 Comparing algorithms on problem instances with a scalable parameter
图 9.4 用可伸缩参数比较问题实例上的算法

Comparing比较; 对比; 与…类似; 表明…与…相似; 将…比作; compare的现在分词
algorithms算法; 计算程序; algorithm的复数
instances例子; 事例; 实例; 举…为例; instance的第三人称单数和复数
scalable可攀登的；可去鳞的；可称量的

Furthermore, there can be few available sets of real data, and these data may be commercially sensitive and therefore diffiffifficult to publish and to allow others to compare.
此外，可用的真实数据可能很少，而且这些数据可能具有商业敏感性，因此很难发布和允许其他人进行比较。
Last, but not least, there might be so many applicationspecifific aspects involved that the results are hard to generalise.
最后，但并非最不重要的是，可能涉及太多应用程序特定的方面，结果很难概括。
Despite these drawbacks it remains highly relevant to tackle real-world problems as the proof of the pudding is in the eating!
尽管有这些缺点，它仍然高度相关的解决现实世界的问题，因为布丁的证据是在吃！

and these形形色色
may be多半
commercially商业上；通商上
and therefore因此，为此
so many那么多的; 很多; 和…一样多的; 全都是
drawbacks缺点; 不利条件; drawback的复数
tackle应付，处理，解决; 与某人交涉; 向某人提起; 抢球; 抢断; 抢截; 铲断; 抢断球，阻截铲球; 擒抱摔倒，阻截; 阻截队员; 用具; 体育器械; 渔具
real-world存在于现实世界的; 真实的
pudding甜食，甜点; 热布丁糕; 热布丁饼

9.4 Example Applications
9.4示例应用
As mentioned in the introduction to this chapter, instead of presenting two case studies with implementation details, we next describe examples of good and bad practice, in order to illustrate some of our points.
正如本章导言中所提到的，为了说明我们的一些观点，我们接下来将描述好的和坏的实践的例子，而不是用实现细节来展示两个案例研究。

mentioned提到; 写到; 说到; mention的过去分词和过去式
instead of代替; 作为…的替换
presenting把…交给; 颁发; 授予; 提出; 提交; 展现，显示，表现; present的现在分词
case studies个案研究; 专题研究; 案例研究; case study的复数
examples实例; 例证; 例子; 典型; 范例; 样品; 榜样; 楷模; 模范; example的复数
in order to为了;以便;目的在于; 要想……
illustrate加插图于; 给做图表; 说明，解释; 表明…真实; 显示…存在
some of中的一些; 其中一些; 其中的一些

9.4.1 Bad Practice
9.4.1不良行为
This section shows a hypothetical example of an experimental study following the template that can be found in many EC publications.¹ In this imaginary case a researcher has invented a new EA feature, e.g., “tricky mutation”, and assessed the value of this new feature by running a standard GA and “tricky GA” 20 times independently on each of 10 objective functions chosen from the literature. The outcomes of these experiments proved tricky GA better on seven, equal on one, and worse on two objective functions in terms of SR. On this basis it was concluded that the new feature is indeed valuable.
本节展示了一个实验研究的假设示例，该模板可在许多EC出版物中找到。4在这个假想的案例中，一位研究人员发明了一个新的EA特性，例如“棘手变异”，并通过在每个模型上独立运行标准GA和“棘手GA”20次来评估这个新特性的价值从文献中选出10个目标函数。实验结果表明，复杂遗传算法在七个目标函数上更好，在一个目标函数上相等，在两个目标函数上较差，在此基础上得出结论，新的特征确实有价值。

hypothetical假设的; 假定的
imaginary想象中的; 幻想的; 虚构的
tricky难办的; 难对付的; 狡猾的; 诡计多端的
mutation变异，突变; 转变，改变
in terms of在…方面;从…角度看;根据…来说

The main question here is what did we, the EC community, learn from this experience? We did learn a new feature (tricky mutation) and obtained some indication that it might be a promising idea to try in a GA. This can of course justify publishing a paper reporting this; however, there are also many things that we did not learn here, including:
这里的主要问题是，我们，欧盟共同体，从这次经历中学到了什么？我们确实学到了一个新特性（棘手的变异），并得到一些迹象表明，在GA中尝试这一点可能是一个很有前途的想法。这当然可以证明发表一篇报道这一点的论文是正当的；但是，还有许多事情我们在这里没有学到，包括：

learn from学习
tricky难办的; 难对付的; 狡猾的; 诡计多端的
mutation变异，突变; 转变，改变
indication表明; 标示; 显示; 象征
promising有希望的; 有前途的; 有出息的; 许诺; 承诺; 答应; 保证; 使很可能; 预示; promise的现在分词
of course当然，自然; 当然; 当然不
justify证明…正确; 对…作出解释; 为…辩解; 调整使全行排满; 使每行排齐; 使齐行
there are有; 可数名词的复数形式; 许多重要事情要谈
did not没有; 未; 的缩写

9 Working with Evolutionary Algorithms
9使用进化算法

• How relevant are these results, e.g., are the test functions typical of realworld problems, or important only from an academic perspective?
• What would have happened if a different performance metric had been used, or if the runs had been ended sooner, or later?
• What is the scope of claims about the superiority of the tricky GA?
• Is there a property distinguishing the seven good and two bad functions?
• Are these results generalisable? Alternatively, do some features of the tricky GA make it applicable for other specific problems, and if so which?
• How sensitive are these results to changes in the algorithm’s parameters?
• Are the performance differences as measured here statistically significant, or can they be just artifacts caused by random effects?
• 这些结果的相关性有多大，例如，测试函数是典型的现实世界问题，还是仅仅从学术角度来看重要？
•如果使用了不同的性能指标，或者运行早或晚结束，会发生什么情况？
•关于棘手GA优越性的主张范围是什么？
•是否有区分七个好功能和两个坏功能的属性？
•这些结果是否具有普遍性？或者，棘手的遗传算法的某些特性是否使其适用于其他特定问题，如果是的话，是哪一个？
•这些结果对算法参数的变化有多敏感？
•此处测量的性能差异是否具有统计学意义，或者是否只是随机效应造成的伪影？

The next example explicitly addresses some of these issues and therefore forms a showcase for a better, albeit still not perfect, practice.
下一个例子明确地解决了其中的一些问题，因此形成了一个更好的（尽管还不完美）实践的展示。

superiority优越; 优势; 优越感; 神气活现的样子; 盛气凌人的行为
tricky难办的; 难对付的; 狡猾的; 诡计多端的
distinguishing区分; 辨别; 分清; 成为…的特征; 使具有…的特色; 使有别于; 看清; 认出; 听出; distinguish的现在分词
make it到达; 及时赶上参加; 成功
if so如果这样; 要是那样的话
forms类型; 种类; 形式; 外表; 样子; 表格; 出现，产生; 形成; 成形，组成; 制作; form的第三人称单数和复数
showcase展示的场合; 玻璃柜台，玻璃陈列柜; 使展现; 在玻璃橱窗陈列
albeit尽管; 虽然

9.4.2 Better Practice
9.4.2更好的实践
A better example of how to evaluate the behaviour of a new algorithm takes into account questions such as:
如何评估新算法的行为的一个更好的例子考虑了以下问题：
• What type of problem am I trying to solve?
• What would be a desirable property of an algorithm for this type of problem, for example: speed of finding good solutions, reliably locating good solutions, or occasional brilliance?
• What methods currently exist for this problem, and why am I trying to make a new one, i.e., when do they not t perform well?
• 我想解决什么样的问题？
•对于这类问题，算法的理想特性是什么，例如：找到好的解决方案的速度、可靠地找到好的解决方案，或者偶尔会有辉煌？
•目前有哪些方法可以解决这个问题，为什么我要尝试制定一个新的方法，即，它们在什么时候表现不佳？

brilliance透明度，亮度，光彩; 缩写词为Bril或BRIL

After considering these issues, a particular problem type can be chosen, a careful set of experiments can be designed, and the necessary data to collect can be identifified. A typical process might proceed along the following lines:
在考虑了这些问题之后，可以选择一个特定的问题类型，设计一组仔细的实验，并确定需要收集的数据。一个典型的过程可能会沿着以下路线进行：
•inventing a new EA (xEA) for solving problem X
• identifying three other EAs and a traditional benchmark heuristic for problem X in the literature
• asking when and why xEA could be better than the other four methods
• obtaining a problem instance generator for problem X with two parameters: n (problem size) and k (some problem-specific indicator)
• selecting five values for k and five values for n • generating 100 random problem instances for all 25 combinations
• executing all algorithms on each instance 100 times (the benchmark heuristic is also stochastic)
• recording AES, SR, and MBF values and standard deviations (not for SR)
• identifying appropriate tests based on the data and assessing the statistical
significance of results
• putting the program code and the instances on the Web
•发明一种新的求解问题X的EA（xEA）
•确定其他三个EA和文献中prob lem X的传统基准启发式
•询问xEA何时以及为什么比其他四种方法更好
•使用两个参数为问题X获取问题实例生成器：n（问题大小）和k（某些特定于问题的指标）
•为k选择5个值，为n选择5个值•为所有25个组合生成100个随机问题实例
•在每个实例上执行所有算法100次（基准heuris tic也是随机的）
•记录AES、SR和MBF值和标准偏差（不适用于SR）
•根据数据确定适当的测试，并评估统计数据
结果的重要性
•将程序代码和实例放到Web上

set of的; 集合; 释……的集合
proceed along沿着…前进
inventing发明; 创造; 编造; 捏造; 虚构; invent的现在分词
heuristic启发式的
for all虽然;尽管

The advantages of this template with respect to the one in the previous example are numerous:
• The results can be arranged in 3D: that is, as a performance landscape over the (n, k) plane with special attention to the effect of n on scale-up.
• The niche for xEA can be identified, e.g., weak with respect to other algorithms for (n, k) combinations of type 1, strong for (n, k) combinations of type 2, comparable otherwise. Thus the ‘when’ question can be answered.
• Analysing the specific features and the niches of each algorithm can shed light on the ‘why’ question.
• A lot of knowledge has been collected about problem X and its solvers.
• Generalisable results are achieved, or at least claims with well-identified
scope based on solid data.
• Reproduction of the results, and further research elsewhere, is facilitated.
与前一个示例相比，此模板的优点有很多：
•结果可以以三维形式排列：即，作为（n，k）平面上的性能景观，特别注意n对放大的影响。
•xEA的利基可以确定，例如，相对于1型（n，k）组合的其他算法法则，xEA的利基较弱；2型的（n，k）组合，其利基较强，否则可与之相比。因此，“何时”的问题可以得到回答。
•分析每种算法的具体特征和利基，可以阐明“为什么”问题。
•收集了大量关于问题X及其解决方案的知识。
•取得了一般性的结果，或至少得到了识别良好的索赔
基于可靠数据的范围。
•有助于结果的复制和其他地方的进一步研究。

with respect to关于，至于（英国英语中也用in respect of）
the one【电影】救世主; 平行歼灭战
that is即; 用于纠正之前说过的内容
scale-up按比例增加；扩大
strong for对…特别偏爱的，特别注重…的
shed light on阐明; 解释; 将…弄明白; 照亮
A lot of许多
has been已经

For exercises and recommended reading for this chapter, please visit www.evolutionarycomputation.org.

---

1. 作者承认，他们自己的一些论文也遵循这个模板。 ↩︎